The Asymptote Principle
Where the idea comes from, how Asympt turns it into one personal priority, and where its limits are.
Abstract
There is more health advice than anyone can act on. What is missing is not information but order. The Asymptote Principle is a way of ordering it: every health behaviour follows a curve of diminishing returns, and the most useful next step for a person is on the curve where they have the most left to gain. This paper sets out where the idea comes from, how Asympt turns it into one personal priority, and where its limits are.
1. The Preston curve
In 1975 the demographer Samuel H. Preston plotted national income against life expectancy and found a characteristic shape. At low incomes, small gains in wealth bought large gains in life, through sanitation, food and basic care. Beyond a point, more wealth bought less and less. The curve flattened towards a ceiling it never reaches.
Preston was describing countries, not people. But the shape turns up almost everywhere in health: exercise, sleep, diet, social contact. The first steps are large; each later step is smaller.
2. From countries to people
If each behaviour follows such a curve, then what to do first differs from person to person. Someone who never exercises and someone who runs five days a week sit at opposite ends of the same curve; the same advice is worth a great deal to one and almost nothing to the other. In 2010, Mold and colleagues made the same argument for clinical medicine: because of diminishing returns, the order of interventions matters as much as the list.
Generic advice ignores position. “Eat more vegetables” is good advice, but not the first thing for someone sleeping five hours a night.
3. The principle
For any health behaviour, the benefit of more effort follows a saturating curve towards a maximum that is approached but never reached. The best next step for a person is on the curve where their own answers leave the most to gain.
Three things follow. The value of advice depends on who receives it. The largest gains are near the bottom of a curve, which is unfashionable in a culture of optimisation but follows from the shape. And there is a natural order: climb the curve with the most left in it until it flattens, then take the next.
4. How Asympt turns it into one priority
One outcome per pillar
Each pillar is scored against one countable outcome: for Longevity, death from any cause; for Fertility, the chance of pregnancy within a year. That keeps every curve testable against something that happens to people, rather than against a questionnaire score.
Curves from the published evidence
Each Longevity domain (exercise, sleep, nutrition, social connection) rests on dose-response studies: how risk changes with the amount of a behaviour. Where studies disagree they are pooled, as a meta-analysis would, rather than one being chosen.
Ranking by risk removed
Within a domain, the risk ratios of a visitor's answers are combined. Domains are then ranked by the share of the visitor's own risk that reaching the domain's best answers would remove. A narrow curve no longer looks as steep as a wide one simply because the visitor sits at its bottom.
One habit, aimed at the weakest lever
The plan follows the answer within the top domain that leaves the most risk: strength training for someone who already walks, a TV off-time for someone who moves but watches five hours a day.
Plans built on habit research
How a habit is built has its own evidence, separate from how much it matters. Asympt's plans use an if-then plan the visitor writes (the best-supported technique), a minimum for bad days, a written tick, and an honest timescale: habits often take around two months, with anywhere from about six weeks to over three months being normal, and one missed day does not undo them. Visitors who ask can get four short check-ins by email, at weeks 4, 9, 13 and 15, each asking one question: does it now happen without you having to think about it? Then the check-ins stop.
What else a habit is linked to
Next to the ranking, each habit shows what it is also linked to beyond death from any cause: dementia, heart disease and stroke, type 2 diabetes, and depression. Each link carries a grade for how strong the evidence is, not how large the effect. These links are not scored and never change the ranking.
5. What the evidence can and can't tell you
Almost every curve in Asympt comes from cohort studies: researchers ask a large group of people about their habits, then follow them for years and count what happens to them. It is the best evidence there is for most of these questions, and it has real weaknesses. Both matter for how you read a result.
Why cohort studies are used
- They follow real people for a long timeHundreds of thousands of people, for 10 to 20 years or more. Trials rarely run that long, and rarely count deaths.
- They show the shapeHow much is enough, and where more stops helping. Asympt is built on that shape.
- They answer what trials can’tNo one can be assigned to smoke, or to be lonely, for 20 years.
- They can be pooledWhen many cohorts in different countries agree, a meta-analysis can combine them.
What can go wrong
- Other differencesPeople who exercise also tend to smoke less, eat better and earn more. Studies adjust for what they measured, not for what they didn’t.
- Illness changes habits firstPeople who are becoming ill move less, sleep differently and see fewer people, years before a diagnosis.
- People misjudge their own habitsMost answers are self-reported, in the studies as on Asympt’s sheet.
- Measured onceA habit reported at the start may have changed over 20 years.
- Who takes partVolunteers tend to be healthier than average. UK Biobank, behind several curves, is a known example.
- A difference between groups is not a promise“People who do X have 20% lower risk” is not the same as “if you start X, your risk falls by 20%”.
What Asympt does about it
- Pooled, adjusted, dose-response analyses firstAdjusted for the main confounders, and checked for agreement across cohorts.
- Where trials exist, they winVitamin D and omega-3 supplements looked protective in cohorts; trials found no benefit. That is why they sit under “Tested, no benefit shown”.
- Illness is kept out where possibleAnalyses that leave out the first years of follow-up, or that measured habits in midlife, are preferred.
- One cohort is not enough to rankSomething resting on a single cohort is listed as popular, not ranked, rather than scored.
- Each curve says what it can’t showBehind “Why this curve?”, next to its numbers.
- A whole can be scored while the plan aims at a partUltra-processed food is scored as a whole, but the plan starts with sugary drinks and processed meat: when studies split it into types, that is where the risk shows up most consistently.
- The ranking compares, it does not predictIt shows which habit has the most to gain, not anyone’s chance of dying.
Lifespan and healthspan
Asympt ranks by risk of death, but the evidence points the same way for years lived in good health. In two large studies, from the US and from Europe, people with the healthiest combination of habits lived about 8 to 11 more years free of major chronic diseases, such as heart disease, type 2 diabetes and cancer, than people with none. They also spent a smaller share of later life with disease: for women in the US study, about 16 in 100 of their years from age 50, against about 25 in 100. A third study found fewer years lived with Alzheimer's dementia.
Three limits come with this. First, the combinations studied were not smoking, being active, a healthy weight, moderate drinking and, in one, diet quality. Only exercise and smoking match Asympt's curves directly; sleep, social contact and ultra-processed food were not part of them. Second, the one study that compared single habits put a healthy weight first, activity second and not smoking third. So these years can't be claimed for Asympt's curves, and they don't change the order. Third, a healthy life postpones disease more than it prevents it: in a Dutch cohort, more than 90 in 100 people developed a chronic disease in their lifetime, even without the main risk factors.
6. Truth mapping
Health advice is sorted by attention: what is talked about most feels most important. Truth mapping sorts it by evidence instead, and shows the two side by side, so the gap between them is visible.
Every habit, trend and supplement Asympt considers ends up in exactly one place:
- Ranked on a curveEnough dose-response evidence to score, and so to compete for your first habit.
- Comes firstSmoking, and drinking seven or more drinks a week: named before any ranking, because the evidence against them is so clear.
- Worth having checkedThings a habit can't fix but a measurement can, such as blood pressure.
- Popular, not ranked: too little evidenceTalked about widely, with too little evidence to put on a curve. Shown with the reason.
- Popular, not ranked: tested, no benefit shownTrials have looked, and found no benefit.
- Not a habit: a measure or a claimTalked about as if it were something to do, but it is a measurement, such as visceral fat, or a claim, such as blue zones. Shown with where the related habit is ranked.
Nothing is quietly scored and nothing is quietly left out. The truth map below puts the two side by side. Each row is where something sits in Asympt; its place along the row is how much it is talked about, measured by a dated trend scan with a published method, so attention is measured rather than guessed.
Ranked on a curve 9
Comes first, or worth checking 3
Popular, too little evidence 22
Tested, no benefit shown 3
Not a habit: a measure or a claim 3
All 40, as a list
- protein: Popular, too little evidence. PN-06.
- sleep: Ranked on a curve. Ranked (SLD-01, SLD-03).
- vitamin D: Tested, no benefit shown. PN-15: trials found no benefit.
- peptides: Popular, too little evidence. PN-17.
- creatine: Popular, too little evidence. PN-18.
- coffee: Popular, too little evidence. PN-05.
- gut health: Popular, too little evidence. PN-19.
- alcohol: Comes first, or worth checking. Comes first above seven drinks a week.
- strength training: Ranked on a curve. Ranked (SX-01).
- blood pressure: Comes first, or worth checking. Worth having checked (LDM-03).
- tai chi: Popular, too little evidence. PN-20: cuts falls in older adults in trials; no trial has measured risk of death.
- omega-3: Tested, no benefit shown. PN-16: trials found no benefit.
- intermittent fasting: Popular, too little evidence. PN-04.
- vegetables: Ranked on a curve. Ranked as vegetables and fruit (NUD-02).
- red light therapy: Popular, too little evidence. PN-21.
- sitting: Ranked on a curve. Ranked as hours of TV (ADD-02).
- metformin: Popular, too little evidence. PN-12.
- NAD+: Popular, too little evidence. PN-12.
- mitochondria: Not a habit: a measure or a claim. A part of the cell, not a habit; we found no supplement sold for it that has been tested against risk of death.
- sauna: Popular, too little evidence. PN-02.
- visceral fat: Not a habit: a measure or a claim. A measure, not a habit: Asympt shows your weight but does not score it (BMD-01).
- multivitamin: Tested, no benefit shown. PN-14: trials found no benefit.
- Bryan Johnson Blueprint: Popular, too little evidence. One person's protocol, never tested as a whole; its parts sit in the ledger separately.
- tadalafil: Popular, too little evidence. PN-23: the longevity claim rests on health records of men prescribed it, not on trials.
- biological age test: Popular, too little evidence. PN-24: no study shows that taking one changes anything.
- psilocybin: Popular, too little evidence. PN-23: a controlled drug; its trials are small, and long-term safety has barely been studied.
- blue zones: Not a habit: a measure or a claim. A claim about places, not a habit; whether the ages behind it are reliable is disputed in published research (E274, E275).
- VO2 max: Ranked on a curve. Ranked as aerobic activity (SX-02, ADD-01): a measure of fitness that rises with the activity on your curves.
- methylene blue: Popular, too little evidence. PN-23: can interact dangerously with common antidepressants.
- daily steps: Ranked on a curve. Ranked, converted to minutes (EXD-02).
- loneliness: Ranked on a curve. Ranked as seeing friends or family (SOD-01); loneliness itself did not hold up for mortality once contact was accounted for (SX-06).
- cold plunge: Popular, too little evidence. PN-03.
- epigenetic reprogramming: Popular, too little evidence. Tested in mice only, with safety questions open; nothing is available to take (E276).
- quit smoking: Comes first, or worth checking. Comes first.
- full body MRI: Popular, too little evidence. PN-24: no guideline recommends scanning healthy people.
- zone 2 training: Ranked on a curve. Ranked as aerobic activity (SX-02).
- continuous glucose monitor: Popular, too little evidence. PN-13.
- rapamycin: Popular, too little evidence. PN-12.
- breathwork: Popular, too little evidence. PN-22.
- ultra-processed food: Ranked on a curve. Ranked (NUD-03).
Of the 40 terms in the latest scan, 9 are ranked on a curve, 3 come first or are worth having checked, 22 have too little evidence to rank, 3 were tested and showed no benefit, and 3 are not habits at all. Of the ten most talked about, 2 are ranked on a curve (sleep and strength training). Every term has been assessed.
The idea has ancestors. David McCandless's Snake Oil Supplements? plotted the evidence for supplements against their popularity. Evidence and gap maps, developed by 3ie and the Campbell Collaboration, show where evidence exists and where it is missing; many are now living maps, kept current with the help of machine learning, and Cochrane is piloting them for global health. AI search tools such as Consensus and scite weigh how papers answer a question. What Asympt adds is the combination: attention and evidence on one map, for everyday habits, every number traced to its place in a paper and signed off by a doctor, and the result turned into one person's first habit.
7. AI traceability: open evidence, closed engine
Asympt uses AI to find and extract numbers from papers. AI is fast and makes mistakes, often confident ones. So every number carries its trail:
- The paperEvery source has a permanent identifier (DOI, OpenAlex) and is resolved before it enters the ledger.
- The numberTaken verbatim, with where it is in the paper: table, figure or page.
- Who extracted itNamed, including the AI system and the session, when AI did it.
- An independent second readThe number is checked against the paper again, separately from the extraction.
- The doctor’s sign-offA doctor checks the numbers and makes every decision about how they are used, dated.
- Its status, shownNot yet verified, checked against the paper, or doctor-verified: next to the number, so a step not yet taken is visible too.
AI never decides what is ranked, how numbers are used, or whether a curve is signed off. That responsibility stays with the doctor. This follows the 2025 joint position of Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence on AI in evidence synthesis: the people doing the synthesis stay responsible for it, and report their use of AI openly. The second reads have caught AI errors before they reached a curve, which is why they exist.
Every curve shows its source and how confident the evidence is.
What stays private is the engine: how the curves are combined and calibrated into a result. The evidence is open; the arithmetic that turns it into your ranking is not.
8. Limits
- Observational evidenceAlmost every curve comes from following people over time, not from trials. Section 5 sets out what that can and can’t show, and what Asympt does about it.
- Self-reportAnswers are estimates, and people misjudge their own habits.
- Combining domainsRisk ratios are combined as if the behaviours were independent. They are not entirely: exercise improves sleep, and social contact and activity overlap. The studies’ own adjustments reduce this, but do not remove it.
- Overlap within a domainWithin nutrition, ultra-processed food and overall diet quality overlap, so some of the same risk may be counted twice. In some large studies the link with ultra-processed food weakens once diet quality is accounted for.
- Relative, not absolute, riskThe ranking compares shares of a person’s own risk. It does not tell anyone their chance of dying.
- Habit evidence is about habitsThe plan techniques are shown to help people keep a behaviour going, not to lower risk of death.
- Not for people who are illThe evidence is drawn, where possible, from people healthy at the start. Asympt is for healthy adults and is not medical advice.
These limits are real. The alternative, a list of everything in no order, has worse ones.
9. Conclusion
Preston showed that the returns on investment in health flatten. Applied to one person, that shape says something practical: start where you have the most left to gain, build one habit until it runs on its own, then look again. Asympt's contribution is not new science. It is structure applied to the science that already exists, with every step of it open to inspection.
Found a problem?
If a number doesn't match its paper, a study is missing or misread, or the method has a flaw, report it. Every report is read by the doctor who signs off the evidence, and a correction changes the ledger, with its date.
Selected references
- Preston SH. The changing relation between mortality and level of economic development. Population Studies 1975;29(2):231–248. doi
- Mold JW, Hamm RM, McCarthy LH. The law of diminishing returns in clinical medicine: how much risk reduction is enough? J Am Board Fam Med 2010;23(3):371–375. doi
- García L et al. Non-occupational physical activity and risk of cardiovascular disease, cancer and mortality outcomes. Br J Sports Med 2023. doi
- Lee DH et al. Long-term leisure-time physical activity intensity and all-cause and cause-specific mortality. Circulation 2022;146:523–534. doi
- Shen X et al. Nighttime sleep duration, 24-hour sleep duration and risk of all-cause mortality. Sci Rep 2016. doi
- Wang DD et al. Fruit and vegetable intake and mortality. Circulation 2021. doi
- Dehghan M et al. Association of unprocessed, minimally processed and ultraprocessed foods with mortality (PURE). Am J Clin Nutr 2023. doi
- Fang Z et al. Association of ultra-processed food consumption with all cause and cause specific mortality. BMJ 2024. doi
- Foster HME et al. Social connection and mortality in UK Biobank. BMC Med 2023. doi
- Li Y et al. Healthy lifestyle and life expectancy free of cancer, cardiovascular disease, and type 2 diabetes: prospective cohort study. BMJ 2020. doi
- Nyberg ST et al. Association of healthy lifestyle with years lived without major chronic diseases. JAMA Intern Med 2020. doi
- Dhana K et al. Healthy lifestyle and life expectancy with and without Alzheimer’s dementia: population based cohort study. BMJ 2022. doi
- Licher S et al. Lifetime risk and multimorbidity of non-communicable diseases and disease-free life expectancy in the general population: a population-based cohort study. PLoS Med 2019. doi
- Lally P et al. How are habits formed: modelling habit formation in the real world. Eur J Soc Psychol 2010. doi
- Gollwitzer PM, Sheeran P. Implementation intentions and goal achievement. Adv Exp Soc Psychol 2006. doi
- Flemyng E et al. Position statement on artificial intelligence (AI) use in evidence synthesis across Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence 2025. Environ Evid 2025. doi
- Kozakiewicz T et al. Living evidence and gap maps: a scoping review of automation and living mode parameters reported in 44 digital interfaces and their documentation. J Clin Epidemiol 2026. doi
- Snilstveit B et al. Evidence & gap maps: a tool for promoting evidence informed policy and strategic research agendas. J Clin Epidemiol 2016. doi
The full evidence for every curve, with locators and verification, is in the open ledger behind “Why this curve?” on your results.